EP0203051A1

EP0203051A1 - Apparatus for determining the degree of variation of a feature in a region of an image that is divided into discrete picture elements

Info

Publication number: EP0203051A1
Application number: EP86850180A
Authority: EP
Inventors: Hans Knutsson; Gösta Granlund
Original assignee: ContextVision AB
Current assignee: ContextVision AB
Priority date: 1985-05-23
Filing date: 1986-05-21
Publication date: 1986-11-26
Also published as: US4747151A; SE8502569L; SE448124B; JPS61272885A; NO861874L; SE8502569D0

Abstract

The invention concerns an apparatus for determining the degree of variation of a feature in a region of an image that is divided into discrete picture elements, said feature being represented by complex valued signals, one for each picture element, the signal phase representing the feature type and the signal amplitude representing the certainty in the feature asssertion: The apparatus includes a unit (3) for providing the complex valued signals (8) within the region and complex valued multiplication factor signals (9) corresponding to said signals. In a first summation unit (4), a first sum signal is generated by the magnitude products of the signals (8) and corresponding multiplication factor signals (9), and a second sum signal is generated by the complex valued scalar products of the signals (8) and corresponding complex conjugate multiplication factor signals. In a second summation unit (5), a third sum signal is generated by the signals (8) that are weighted with the magnitude of corresponding multiplication factor signals (9), and a fourth sum signal is generated by the complex conjugate multiplication factor signals (9) weighted with a magnitude of corresponding signals (8). A norming unit (6) is provided for norming the output signals from the summation units in a predetermined way (Figure 2).

Description

The present invention concerns an apparatus for detecting fast and reliably the degree of variation of a feature in a region of an image that is divided into discrete picture elements.

BACKGROUND OF THE INVENTION

When interpreting the contents of digitized images the detection of regions with homogeneity in characterizing features is an important step of the analysis. These regions are typically characterized in that one or several features within said regions are relatively constant or vary in a continuous and predictable way. Such continuously varying features may for instance be the curvature of lines and edges of an image. It could also be a matter of gradual shadings of the lighting of an image, or the variation in the resolution of details as a function of the distance in a perspective image.
One on-going problem has been that the characterization of the variation means differentiation, which is an operation that is specially sensitive to noise. It has therefore been difficult to distinguish between low level noise in the picture and characterizing features that show a stable variation. The present invention concerns an apparatus by means of which it is possible to distinguish regions that have an even, continuous and predetermined variation of a feature from those regions that show an uneven, random variation.
The present invention originates from research within areas concerned with computerized image analysis. The literature within this area describes various and different algorithms, which relate to those problems that are solved by this invention. The signal processing performed by the present invention could, as well as these algorithms, in principle be performed by a general computer. All such implementations do have the drawback that the processing is not realized as hardware, but require a program of some kind. Consequently the processing times involved become unreasonably long for all applications of interest.

SUMMARY OF THE INVENTION

The present invention thus discloses an apparatus that with high speed and reliability provides such a detection of the degree of variation of a feature in a region of an image that is divided into discrete picture elements.
The features characterizing the invention will appear from the accompanying patent claims.
It has proven that such an apparatus with the special applications referred to herein, can be thousands of times faster than a general computer.
The invention will be further described below in regard to an apparatus for the examination of a local region of an image that is divided into discrete picture elements. The values of the picture elements can be represented in either digital or analogous form, depending on the implementation of the system. A complete system for the examination of a whole image can either contain a number of such units working in parallel with different sub-regions of the image, or one single unit that successively analyses each sub-region of the image. These sub-regions can partially overlap each other. It is here assumed that the features to be observed are represented in the form of complex valued signals or two-dimensional vectors, one for each picture element. These signals or vectors represent one feature, for example dominant orientation. Each vector is of such a nature that its direction describes the dominant orientation of a structure in the image, while the length of the vector represents the certainty in the assertion of this direction. This vector representation of the image can be obtained for example through the conversion of the original, digitized image, in which each picture element can be allocated a grey scale level or an intensity level combined with a color code, according to the principles disclosed in the "IEEE TRANSACTIONS ON COMMUNICATIONS", VOL COM-31, No. 3, March 1983.
The apparatus, by means of which the present invention solves the above indicated problems, can be characterized by a combination of four main components. One or several units provide image data within the viewed region and complex valued multiplication factor signals that describe the predetermined degree of variation for said corresponding region. A first summation unit correlates the vectors within the viewed region in a predetermined way to said multiplication factors. A second summation unit sums up current complex valued image data and multiplication factors in a predetermined way. The unit then norms the output signals from the first summation unit and the second summation unit, respectively, in a predetermined way. The complex valued output signal, which thus is obtained, constitutes a measure of the correspondence between the degree of variation determined by the multiplication factors and the actual variation of said feature within the detected region. The phase of the complex valued output signal constitutes a qualitative assertion about the feature within the region, while the magnitude of the signal constitutes a measure of the certainty in the assertion, that is, the greater the magnitude, the more certain the assertion.

BRIEF DESCRIPTION OF THE DRAWING

The invention will be further described below with reference to the accompanying drawing, in which:

Fig. 1 illustrates the problem situation in question,
Fig. 2 shows a block diagram of the apparatus according to the present invention,
Fig. 3 shows a more detailed block diagram of an addressing unit with memories for picture data and multiplication factors,
Fig. 4 shows a more detailed block diagram of the first summation unit for correlative calculation of Fig. 2,
Fig. 5 shows a more detailed block diagram of the second summation unit for summing up the complex valued signals of Fig. 2,
Fig. 6 shows a more detailed illustration of the norming unit of Fig. 2,
Fig. 7 shows two examples of "masks" describing two different predetermined variation degrees and including said multiplication factors,
Figs. 8 - 9 show parts of alternative, analogous embodiments of the present invention,
Fig. 10 shows some further developments of the invention, and
Fig. 11 shows an alternative embodiment of the norming unit of Fig. 2.

DESCRIPTION OF THE PREFERRED EMBODIMENT.

Fig. 1 illustrates the current problem situation. A region 1 with an evenly varying feature, representing for instance a certain structure orientation within the region, is characterized by complex valued signals or vectors with a direction that varies in a continuous way. An area 2 with a randomly varying or a non-homogeneous feature, means that the characterizing vectors have an orientation that varies in a noisy way. The problem situation is such that it is desirable, in a quick and reliable way, to be able to distinguish regions of type 1 from regions of type 2. It is thus desirable to be able to measure the degree of variation of a feature within a region around a picture element.
Fig. 2 shows a block diagram of an apparatus according to the present invention. The Figure shows a unit 3 that provides complex valued multiplication factor signals as well as image data for a region corresponding to a sub-region or a "window" of a digitized image, which is converted according to the above, within which region a measurement is to be carried out. Typically such a region contains 11 x 11 two-dimensional vectors. The values of the vector components within this region are detected by two different calculating units 4 and 5. The output results from these two calculating units are then compared in a norming unit 6, where a normed result is obtained. This norming unit 6 delivers a complex valued signal 7 that characterizes the degree of variation of the feature within the region in question.
An embodiment of the apparatus according to Fig. 2 will now be described with reference to Figs. 3 - 6.
Fig. 3 shows more in detail the unit 3 of Fig. 2 for providing complex valued multiplication factor signals and picture data for the viewed region. It is here assumed that the different complex valued signals or the vectors occurring within one neighborhood around the "measuring point" of the image, the signal 8, as well as the multiplication factors, the signal 9, are represented by magnitude and phase components, that is in polar format. In a digital representation the magnitude for example can be represented by an integral number between 0 and 255, that is, by one byte. In the same way, the phase angle can also be represented by one byte. Each picture element will therefore in a memory be allocated two bytes which represent the signal of the picture element. If another kind of digital resolution is desired, other storage allocations are of course conceivable. The data can also be represented in analogous form such as voltage or current. An address generator 10 provides from an image data memory 11 current vector values 8 for a neighborhood around the measuring point in the form of a magnitude signal 12 and a phase signal 13. The address generator 10 also provides a corresponding set of complex valued multiplication factors in the form of a magnitude signal 15 and a phase signal 16 from a coefficient memory 14. The units 10, 11 and 14 are each well known in the art and are not objectives of this invention.
Thus, corresponding to each picture element in the viewed region of the image are an actual complex valued signal or vector, as well as a complex multiplication factor. The multiplication factors form together a "mask" that describes a predetermined, idealized variation of the current feature within the region. Examples of such masks are shown in Fig. 7. The purpose of the invention is to measure the correspondence between the actual set of vectors and this hypothetical, idealized variation.
Fig. 4 shows more in detail the structure of the unit for correlating the vectors within the viewed region to said complex valued multiplication factors. The magnitude signal 12 of the picture data memory 11 and the magnitude signal 15 of the coefficient memory 14 are passed to a multiplicator 17. The resulting product components 18 are summed in a summator 19, and the resulting product sum for the viewed neighborhood is obtained at output 20.
The phase signal 13 of the picture data memory 11 and the phase signal 16 of the coefficient memory 14 are passed to an input and an inverting input, respectively, of an adder 21. The obtained phase difference is then taken to a look up table 22 for cosine and a look up table 23 for sine. The values obtained through this procedure are each supplied to a multiplying summator 24 and 25, respectively, where they are multiplied by the signal 18, followed by a summation of the result. This provides product sums 26, 27 for the real and imaginary part, respectively, for the viewed neighborhood.
Fig. 5 shows more in detail the structure of unit 5 of Fig. 2 for the summation of current complex valued image data within the neighborhood as well as complex valued multiplication factors. The phase signal 13 of the picture data memory 11 and the sign inverted phase signal 16 of the coefficient memory 14 are each taken to a corresponding look up table 28 and 30, respectively, for cosine and each to a corresponding look up table 29 and 31, respectively, for sine. The thus obtained values are each passed to a multiplying summator 32, 33, 34 and 35. The magnitude product 18 is present as an additional input signal to each multiplying summator. This provides product sums for the real part, 36 and 38, respectively, and the imaginary part, 37 and 39 respectively, for the viewed neighborhood.
Fig. 6 shows more in detail the norming unit 6 of Fig. 2. The earlier obtained product sums 36, 37 and 38, 39, respectively, for the viewed neighborhood are passed to a unit 40 for a complex valued multiplication of these sums. The real part 41 and the imaginary part 42 of the resulting product are each brought to a dividing unit 43 and 44, respectively, where they are divided by the earlier obtained magnitude product sum 20, to form the values 45 and 46, respectively. These values are subtracted from the earlier obtained product sums 26 and 27 in two adders 47 and 48, respectively, each provided with a sign inverting input. Subsequently, the obtained values are divided in dividing units 49 and 50 by the signal 20 for forming the desired final output signals 51 and 52, respectively. These can then, if desirable, be brought to polar format by means of a rectangular/polar converter 53.
Fig. 7 shows two examples of the multiplication factors that are useful in conjunction with the present invention. Since it has been assumed that the viewed region of the image is formed by a set of 11 x 11 picture elements (of course, also other sizes are conceivable) the multiplication factors as well form a set of 11 x 11 complex valued signals, each factor corresponding to one picture vector within the region.
Fig. 7a shows a "divergence" mask which is suitable for detecting e.g. end points on lines and corners in the image.
Fig. 7b shows a "rotary" mask which is suitable for detecting e.g. radial and circular structures in the image.
Thus, it is understood that the interpretation of the measuring result depends on which mask of complex valued multiplication factors that was used in the measurement.
The earlier described implementation primarily relates to digital technique, whereby the vectors of the picture elements are represented in digital form. However, the indicated functions can advantageously also be performed in an analogous technique. Those structures that become specific for this form of implementation are provided in Figs. 8 and 9.
Fig. 8 shows an analogous variant of the unit 4, and Fig. 9 shows an analogous variant of unit 5.
According to Fig. 8 the first summation unit 4 sums all the real parts and imaginary parts of the signals 8 multiplied by respective multiplication factors that are implemented by resistor networks 90 - 93 or multiplying D/A converters that are controlled by digital control signals 100 - 103. Then the signals are passed further to summators 110 - 113, for instance in the form of feed back operational amplifiers. It is to be noticed that the whole neighborhood around the measuring point (n = 11 x 11 = 121, under the same conditions as above) is processed in parallel, in this embodiment. Multiplications in the networks 90 - 93 can be regarded as a reflection of Ohm's law. The advantage with multiplying D/A converters is that the resulting resistance value corresponding to the multiplication factor is adjustable by a digital control signal. The output signals from the summators 110 - 113 are summed in adders 120, 121 for obtaining signals corresponding to the signals 26, 27.
The magnitude product sum 20 is obtained in a similar way by means of the magnitude signals for the vectors and the multiplication factors for the neighborhood, respectively, the latter being implemented by a network 80 of resistors or multiplying D/A converters. The product signals are summed in a summator 82 for obtaining the signal 20.
According to Fig. 9 the second summation unit 5 sums all the real and imaginary parts, respectively, of the signals 8 with the magnitudes of the multiplication factors implemented by networks 130, 131 of resistors or multiplying D/A converters controlled by control signals. The output signals from respective networks are summed in summators 140, 141. In a similar way, the magnitude signals of the vectors are summed, the vectors being multiplied by the real and imaginary parts, respectively, of the multiplication factors that are implemented by networks 132, 133 of resistors or multiplying D/A converters controlled by control signals. The signals are then each passed further to a summator 142 and 143, respectively, the latter being provided with inverting inputs. From the summators 140 - 143 the earlier mentioned signals 36 - 39 are obtained.
In the embodiment according to Figs. 8 and 9 the signals are assumed to be accessible as real and imaginary parts, as well as magnitude. This provides a particularly simple structure of the apparatus. Thus, it can be desirable to store the picture signals in rectangular form as well as in polar form, or, besides rectangular form, at least also store the vector magnitudes. The resulting additional storage requirement is compensated by the simplified structure of the apparatus. This is in particular applicable if the whole image or greater parts of the image are to be examined in one step by a plurality of apparatuses that work in parallel.
Some further developments of the invention are shown in Fig. 10 for the provision of a somewhat more general apparatus.
Sometimes it is desirable to be able to vary the norming. In the above described embodiments the norming is maximized, and the output signal is independent of the energy of the input signal. In the purpose of further reducing the noise sensitivity of the measurement, a unit 150 can be provided for the dividing units 49, 50 in the norming unit 6. This unit raises the signal 20 to an exponent between 0 and 1. The size of the exponent is preferably adjusted by a control signal 151. The special case, in which the exponent is 0, can also be implemented by omitting the unit 150 and dividing units 49, 50. The special case, in which the exponent is 1, is equivalent to the embodiment of Fig. 6. Thus, in this case the unit 150 can be omitted.
Yet another generalization of the invention is obtained if an adder 160 to the signal 20, which is possibly processed according to the preceding paragraph, adds a reference or base level 161. This is in particular suitable in cases with low input levels since these levels otherwise can render a misleading result in a normal norming. In the special case, when the reference level is 0, the adder 160 can be omitted and the embodiment according to Fig. 6 is recovered. Fig. 11 shows an alternative embodiment of the norming unit 6 of Fig. 2. Since this embodiment in greater parts resembles the embodiment of Fig. 6, the same reference designations have been used, when possible. In this embodiment the dividing units 43, 44 of Fig. 6 have been omitted. Instead a pair of multiplicators 180, 181 have been connected in front of the adders 47, 48. There, the signals 26, 27 are multiplied by the signal 20 before they are passed to the adders. Another difference is that the signals 41, 42 now directly are passed to the adders 47, 48. These adders are, as before, each connected to one of the inputs of a corresponding dividing unit 49, 50. Before the signal 20 is supplied to the second input of these dividing units, it can, in an analogous way to Fig. 10, be supplied to a unit 170 that raises it by a controllable exponent 171, which in this case is between 1 and 2. In the special case when the exponent equals 2, the output signal 51', 52' of the norming unit will be identical with the output signal from unit 6 of Fig. 6. In the other extreme case, when the exponent equals 1, the unit 170 and the units 49, 50 can be omitted. As in Fig. 10, a reference signal 191 in an adder 190 can here as well be added to the output signal from unit 170.
In the embodiments according to Figs. 10 and 11 certain intervals have been indicated for the exponents, by way of an example. It is, however, understood that the invention is not confined to exactly these intervals, and that also other intervals are conceivable.
A person skilled in the art realizes that these described embodiments of the invention can be varied and modified in several ways within the frame of the basic idea of the invention, which is described in the accompanying patent claims.

Claims

1. Apparatus for determining the degree of variation of a feature in a region of an image that is divided into discrete picture elements, said feature being represented by complex valued signals, one for each picture element, the signal phase representing the feature type and the signal amplitude representing the certainty in the feature assertion, characterized by:

one or several units (3; 80, 90 - 93, 130 - 133) for providing said complex valued signals (12, 13) within said region and complex valued multiplication factor signals (15, 16) corresponding to said signals;

a first summation unit (4) for forming a first sum signal (20) from the magnitude products of said complex valued signals (12) and said corresponding multiplication factor signals (15), and for forming a second sum signal (26, 27) from the complex valued scalar products of said complex valued signals (12, 13) an said corresponding multiplication factor signals (15, 16);

a second summation unit (5) for forming a third sum signal (36, 37) from said complex valued signals (12, 13) weighted with the magnitude of said corresponding multiplication factor signals (15), and for forming a fourth sum signal (38, 39) from the complex conjugate of said multiplication factor signals (15, 16) weighted with the magnitude of said corresponding complex valued signals (12); and

a norming unit (6) for norming said second sum signal (26, 27) subtracted by the complex product (41, 42) of said third and fourth sum signals (36, 37; 38, 39) divided by said first sum signal, with respect to the sum of said first sum signal (20) raised to a predetermined exponent (151) and a reference signal (161, k).

2. Apparatus for determining the degree of variation of a feature in a region of an image that is divided into discrete picture elements, said feature being represented by complex valued signals, one for each picture element, the signal phase representing the feature type and the signal amplitude representing the certainty in the feature assertion, characterized by:

a first summation unit (4) for forming a first sum signal (20) from the magnitude products of said complex valued signals (12) and said corresponding multiplication factor signals (15), and for forming a second sum signal (26, 27) from the complex valued scalar products of said complex valued signals (12, 13) and said corresponding multiplication factor signals (15, 16);

a norming unit (6) for norming the product of said second sum signal (26, 27) and said first sum signal (20) subtracted by the complex product (41, 42) of said third and fourth sum signals (36, 37; 38, 39) with respect to the sum of said first sum signal (20) raised to a predetermined exponent (171) and a reference signal (191, k').

3. Apparatus according to claim 1 or 2, characterized in that the first summation unit (4) includes a multiplicator (17) that is connected to a summator (19) for generating the first sum signal (20) (Fig. 4).

4. Apparatus according to any of the preceding claims, characterized in that the first summation unit (4) includes an adder (21) with two inputs (13, 16), one of which (16) is inverting, for adding the phase components of the two complex valued input signals, a unit (22, 23) connected to the adder for forming cosine and sine, respectively, of the output signal of the adder, and two multiplying summators (24, 25), which multiply the cosine and sine signal, respectively, by the output signal (18) of the multiplicator and which summators sum the results for generating the second sum signal (26, 27) (Fig. 4).

5. Apparatus according to any of the preceding claims, characterized in that the second summation unit 5 includes two units (28, 29; 30, 31) for forming cosine and sine of the phase components (13, 16) of the two complex valued input signals, and two multiplying summators (32, 33; 34, 35) each connected to respective cosine/sine signals, which multiplying summators multiply said signals by the output signal (18) of the multiplicator (17) and sum the results for generating the third and fourth sum signal (36, 37; 38, 39), respectively (Fig. 5).

6. Apparatus according to claim 1 or 2, characterized by a network (80) of resistors or controlled, multiplying D/A converters that represent the magnitudes of the complex valued multiplication factor signals, the network being connected to the magnitude components of the complex valued signals (8), and in that the output signals of the network are summed in a summator (82) for generating the first sum signal (20) (Fig. 8).

7. Apparatus according to claim 1, 2 or 6, characterized by two networks (100 - 101) of resistors or controlled, multiplying D/A converters that represent the real and imaginary parts of the complex valued multiplication factor signals (9), the networks being connected to the real and imaginary parts, respectively, of the complex valued signals (8), the output signals of the networks being connected to two summators (110 - 111), by two additional networks (102 - 103) of resistors or controlled, multiplying D/A converters that represent the real and imaginary parts of the complex valued multiplication factor signals (9), the additional two networks being connected to the imaginary and real parts, respectively, of the complex valued signals (8), the output signals of either networks being connected to two additional summators (112 - 113), and by two adders (120, 121) for the summation of the output signals from respective pair of summators, one input of the adder, that belongs to the second pair of summators, being inverting, for generating the second sum signal (26, 27) (Fig. 8).

8. Apparatus according to claims 1, 2, 6 or 7, characterized by two networks (130, 131) of resistors or controlled, multiplying D/A converters, each representing the magnitude components of the complex valued multiplication factor signals, the two networks being connected to the real and imaginary parts, respectively, of the complex valued signals, the output signals of the networks each being connected to a summator (140, 141) for generating the third sum signal (36, 36), by two additional networks (132, 133) of resistors or controlled, multiplying D/A converters representing the real and imaginary parts, respectively, of the multiplication factor signals (9), each network being connected to the magnitude components of the complex valued signals, the output signals from the latter networks each being connected to a summator (142, 143) for generating the fourth sum signal (38, 39) (Fig. 9)

9. Apparatus according to any of claims 1, 3 - 9, characterized in that the norming unit (6) includes a multiplicator (40) for complex valued multiplication of the third sum signal (36, 37) by the fourth sum signal (38, 39) and two dividing units (43, 44) for dividing the real part and imaginary part, respectively, of the output signal of this multiplicator by the first sum signal, and two adders (47, 48) each with two inputs, one of which is inverting, for subtracting the complex valued output signal (45, 46) of the dividing units (43, 44) from the second sum signal (26, 27) (Fig. 6).

10. Apparatus according to any of claims 2 - 9, characterized in that the norming unit (6) includes a multiplicator (40) for complex valued multiplication of the third sum signal (36, 37) by the fourth sum signal (38, 39), two multiplicators for multiplying the real part and imaginary part, respectively, of the second sum signal (26, 27) with the first sum signal, and two adders (47, 48) each with two inputs, one of which is inverting, for subtracting the complex valued output signal (41, 42) of the first mentioned multiplicator (40) from the product of the second sum signal (26, 27) and the first sum signal (Fig. 11).

11. Apparatus according to claim 9 or 10, characterized in that the norming unit (6) includes two dividing units (49, 50) for norming the output signal of the two last mentioned adders (47, 48), with respect to the first sum signal (20) (Fig. 6).

12. Apparatus according to claim 11, characterized by a unit (150, 170) connected to the last mentioned dividing units (49, 50) for raising the first sum signal 20 by a predetermined exponent (150, 171) (Fig. 10, 11).

13. Apparatus according to claim 12, characterized by an exponent (151) between 0 and 1 (Fig. 10).

14. Apparatus according to claim 12, characterized by an exponent (171) between 1 and 2 (Fig. 11).

15. Apparatus according to any of claims 11 - 14, characterized by an adder (160, 190) connected to the last mentioned dividing units (49, 50) for adding a reference signal (161, k; 191, k') to the first sum signal (20) and to the first sum signal raised by the exponent (15₁, 161), respectively (Fig. 10, 11).

16. Apparatus according to any of the preceding claims, characterized in that a rectangular/polar converter (53) is connected to the norming unit (6) for converting the output signal (51, 52; 51', 52') of this unit (6) to polar form (7).